Re: Litmus test for question from Al Viro

[joel@joelfernandes.org]

On Sat, Oct 03, 2020 at 12:08:46PM -0400, joel@joelfernandes.org wrote:
[...] 
> static void code0(struct v_struct* v,spinlock_t* l,int* out_0_r1) {
> 
>         struct v_struct *r1; /* to_free */
> 
>         r1 = NULL;
>         spin_lock(l);
>         if (!smp_load_acquire(&v->b))
>                 r1 = v;
>         v->a = 0;
>         spin_unlock(l);
> 
>   *out_0_r1 = !!r1;
> }
> 
> static void code1(struct v_struct* v,spinlock_t* l,int* out_1_r1) {
> 
>         struct v_struct *r1; /* to_free */
> 
>         r1 = v;
>         if (READ_ONCE(v->a)) {
>                 spin_lock(l);
>                 if (v->a)
>                         r1 = NULL;
>                 smp_store_release(&v->b, 0);
>                 spin_unlock(l);
>         }
> 
>   *out_1_r1 = !!r1;
> }
> 
> Results on both arm64 and x86:
> 
>     Histogram (2 states)
>     19080852:>0:r1=1; 1:r1=0;
>     20919148:>0:r1=0; 1:r1=1;
>     No
>     
>     Witnesses
>     Positive: 0, Negative: 40000000
>     Condition exists (0:r1=1 /\ 1:r1=1) is NOT validated
>     Hash=4a8c15603ffb5ab464195ea39ccd6382
>     Observation AL+test Never 0 40000000
>     Time AL+test 6.24
> 
> I guess I could do an alloc and free of v_struct. However, I just checked for
> whether the to_free in Al's example could ever be NULL for both threads.

Sorry, here I meant "ever be non-NULL".

So basically I was trying to experimentally confirm that to_free could never
be non-NULL in both code0 and code1 threads.